Machine-Vision Detection for Rail-Steel’s Surface Flaws-4300(002)

《Machine-Vision Detection for Rail-Steel’s Surface Flaws-4300(002)》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Proceedingsofthe7thWorldCongressonIntelligentControlandAutomationJune25-27,2008,Chongqing,ChinaMachine-VisionDetectionforRail-Steel’sSurfaceFlawsBasedonQuantumNeuralNetworkXueWangSchoolofElectric&InformationEngineeringChongqingUniversityofScienceandTechnologyChongqing,400042,Chinasaltm

2、aker@163.comYikeTangandPingChengCollegeofMechanicEngineeringChongqingUniversityChongqing,400044,Chinatangyike@cqu.edu.cnAbstract-Conventionaldetectingmethodsbring2)Electriceddydetection:Whenaprobeinsideaelectricdisadvantagesoflow-efficiencyorhigh-falloutrateasforrail-steel’ssurfacef

3、lawsbecauseofitsnon-planarandsophisticatedcontour.Ajournalmachinevisionapproachwaspresented,inwhichimagingmethodandclassifieralgorithmareillustrated.LinerCCDisadaptingtoimagingformovingrail-steel.TheclassifierbasedonQuantumNeutralNetwork(QNN)algorithmcoulddealwiththosesimilarandhardlyd

4、ifferentiatedROIofflaws.Itdiscussedfeaturevectorparametersextractedfromdifferentspaces,moreover,QNN’smodel,multi-levelmotivationfunctionsbasedonSigmoidfunctionandtrainingalgorithmareexpatiatedindetail.Anexperimentaldevicewasdevelopedandtestresultsdemonstratethefeasibilityofthedetection

5、approach.Ithasprovedtheeffectivenessandvalueofproposedmethodinautomaticdetectionforrail-steel’ssurfaceflaws.IndexTerms–Machine-vision,surfaceflaw,rail-steel,QNN.I.INTRODUCTIONRail-steelisakindofimportantproductwhichisindispensablyusedinrailroadtransportationsystem.Qualityoftherail-stee

6、lisnotonlybasicsafeguardbuttheenterprise'scorecompetitiveness.Innerflawsofrail-steelareeasilydetectedinearlymanufacturingtechnologycoursebyX-ray,sogenerallyallrail-steelproductsareacceptableasforinnerflawindexmark.Surfaceflawsdetectionwouldmeettroublesbecausetherail-steel’ssurfaceshape

7、isirregularandcomparativlyrough.Roaringsurfacequalitycontrolneedsinrail-steelcurrentmarket.Inthisarticle,ajournaldetectionapproachbasedonmachinevisionisproposedandtechnologyofimaging,featureparameterextracting,andclassifieralgorithmareillustrated.Detectionmethodsaretakenintotwocatego